AppsFlyer Runs Near-Real-Time, Ultra-Low Latency, High-Throughput Workloads at Scale Using Amazon EKS

Daily, AppsFlyer’s ingress and internal service communication generates around eight hundred billion events. At peak hours, this traffic exceeds 12 million events per second. By adopting a scalable architecture based on Amazon EKS, AppsFlyer can scale infrastructure up and down based on load, paying for only what is used. The company has reduced latency by 30–90 percent, depending on the workload. It now performs version upgrades, configurations, and many other tasks in days or even hours instead of weeks. By binding each Kafka cluster to a different business logic, the company can avoid a single point of failure and tune each cluster for an optimal cost-performance ratio. Using this architecture, AppsFlyer has improved its performance and stability while reducing security risks.

AppsFlyer also enjoys maximum resource efficiency. Using AWS Graviton processors, the company can choose different CPU and storage types based on its needs. In fact, AppsFlyer has reduced its costs by an average of 65 percent thanks to this flexibility. “We improved performance, reduced costs, and did not harm our offering for our customers,” says Gershkovich. “We only improved it.”